Conversation
f-allian
commented
Feb 17, 2026
f-allian
commented
- Adding my NLP blog post from https://casestudiesrcg.blogspot.com/2026/02/a-novel-deep-learning-architecture-for.html
- Any feedback/suggestions are welcome, TIA
| artificially upsampling of rare labels in our approach. Instead, we relied on strict stratified sampling across our | ||
| training, validation, and test splits that mimics the raw dataset's proportions and reduces the model's bias. This | ||
| guarantees that rare technological domains are preserved and adequately represented across all phases of model | ||
| development. A summary of the data splits is shown in Table 1. |
There was a problem hiding this comment.
Could have embedded links to Table 1 here and for other tables/figures even though this table is adjacent and the document isn't too complicated (its a habit I have from using LaTeX to link internally to figures/tables).
| architecture of our multi-label text classification pipeline involves the following three main steps: | ||
|
|
||
| 1. **Preprocessing:** Raw abstracts are tokenised (up to a maximum sequence length of 512 tokens). Each token is mapped to | ||
| a 768-dimensional embedding vector, and the final hidden state of the classification token ([CLS]) is pooled to create a |
There was a problem hiding this comment.
Why is ([CLS]) in both parentheses and square brackets, is it meant to be a hyper-link to something?
| a 768-dimensional embedding vector, and the final hidden state of the classification token ([CLS]) is pooled to create a | ||
| single, dense semantic representation of the entire abstract. | ||
|
|
||
| 2. **Fine-tuning:** The token embeddings are passed into the pre-trained SciBERT layer to perform fine-tuning. [CLS] token |
There was a problem hiding this comment.
As above for [CLS] is it meant to be a hyperlink with the target URL in parentheses missing?
| During validation, we treated the task as a multi-label classification problem, looking at both micro metrics (e.g. | ||
| F1-micro, which favour frequent classes) and macro metrics (e.g. F1-macro, which treat rare niche classes equally) to |
There was a problem hiding this comment.
I've not heard of F1-[micro|macro] before would it be worth linking or citing references that explain these for readers?
|
|
||
|  | ||
| {: style="text-align: center;"} | ||
| ***Figure 3**: Training and evaluation performances of the hierarchical classifier across 20 epochs. (a) Training loss |
There was a problem hiding this comment.
One other thing I just remembered...
Whilst *italics* italicises text and **bold** using the two adjacent in this manner could lead to confusion (I had to think a little about it and only really clicked when I rendered the page).
A solution to make it clearer is to use _italics_ which give the same effect and make the source easier to read.
This is more a matter of personal style, but if you used something markdownlint-cli2 then depending on configuration for rule MD049 it might throw some errors. (Other Markdown linters are available, this is the one I use commonly as a pre-commit hook, I may switch to a Rust based on rumdl in the future).
|
@ns-rse Thanks for your feedback Neil, I've addressed the changes in a new commit. |
2 participants
